Is it possible to to automatically generate vector mpy and/or add instructions from C code (clang) without using intrinsics -- i.e. just from normal for loops? The clang compiler command line options include options for Hexagon vectorization such as -mhvx -mno-hvx. However, these appear to make no difference in the assembler output from the C compiler. Only scalar mpyi instruction are generated. Never vector multiply instruction.
Auto vectorization from clang
Posted: Fri, 2017-10-06 13:59